An AI system says it broke an Enigma message that has stumped cryptographers since 2005.
According to a post on cryptocellar.org, a system described as OpenAI's GPT-6 Astra broke an Enigma-enciphered message that had resisted every previous attempt at decryption. The claim also circulated on a tech discussion forum, where it picked up 63 points and 62 comments before most readers had a chance to check it. The source page does not describe the cryptanalytic method, the size of the key space involved, or how the recovered plaintext was validated. No transcript of the broken message and no technical writeup accompany the announcement.
Enigma traffic has been a proving ground for automated codebreaking since Bletchley Park, and hobbyist projects have spent years chipping away at unsolved wartime messages using brute-force search and statistical attacks, not general-purpose language models. If an off-the-shelf AI system solved one without cipher-specific tooling, that would mark a real shift in what these models can do against classical cryptosystems. But OpenAI has not publicly announced a model called GPT-6, let alone one named Astra, which makes the framing here read more like speculation than confirmed capability.
Until someone publishes the plaintext and the method behind it, file this under good story, not verified result.